Credits on Project 14196 need adjusting

[MeeLee]

Cerdits for project 14196 read about 8-9% low, and probably will need adjusting.

[Joe_H]

The target is to have PPD for various projects within a +/- 10% range, unless it is off for all types of hardware it is probably not going to be changed. There can also be variations between WU's from different runs of a project, they will go for an average over all.

[bruce]

Cerdits for project 14196 read about 8-9% low, and probably will need adjusting.

In addition to what Joe_H has already said, your statement is incomplete since you didn't mention your hardware. For a statement like that to be taken seriously, you would need to mention that P14196 seems to be about 8-9% low when run on a XXXX GPU compared to Pnnnnn which may also be about y% high.

[MeeLee]

All (Non super) RTX GPUS (2060/2070/2080/2080Ti) show between 6-10% lower reading on this project.

[bruce]

Various projects (particularly those with low atom-counts) are not able to keep all the shaders busy on high-shader-count hardware -- particularly if the slot has a reduced bandwidth. What speed are your slots?

The only other choice is the project owner can block all project assignments to high-shader-count GPUs. Unfortunately that's only useful if there are plenty of older GPUs (with fewer shaders) to support the project's performance goals. That's not a permanent solution, though, as older GPUs do get updated.

[MeeLee]

The score is a score, albeit just a bit lower than other projects.
I think it's just not calibrated well.
I saw there was another project (18***, forgot which one) that had even lower PPD rating.
You'd think that when a project runs, and it utilizes the GPU to the fullest (95-100% gpu load), the PPD would be the same...

They do fully utilize faster GPUs.

As far as slots, the same is true for my RTX 2060s that run in PCIE 3.0 1x slots, that used to get 1+M PPD there, and now see 950k PPD, as well as the 2080Tis that are running in PCIE 3.0 full size slots (at 8x).

[toTOW]

Spoiler alert : the GPU load % is not the shader load %, it represents the frontend circuitry (in charge of data management).

The only way to guess shader load is to look at power usage of the chip.